Advanced analytics, where algorithms are applied to historical data to detect and deploy patterns for better business decisions, are moving into a phase of self-service for users. These capabilities have long been the domain of data scientists using their favorite code or solution. Today, this is evolving into methodologies and technologies where business people want and need self-service analytics that takes advantage of these powerful techniques.
If we start this discussion with a business use case, we maximize the chance that the business uses the predictive results for action and decision-making. This means marketing people need targeted lists of prospects to send offers to, and their need is an application that allows for creating that list based on a likelihood of offer interest. Their goal may be to target any customer with an offer that displays a likelihood of purchase greater than 50 percent and an estimated spend of at least $75, subject to a promotional budget of $5,000.
When it comes to advanced analytics, true results deployment is more than data-sharing and populating a dashboard with new attributes.
Likewise, in a manufacturing setting, repair technicians and engineers need a prediction of equipment failure application to guide their decisions when scheduling machine downtime and personnel by taking devices off-line for maintenance. This is critical to minimizing the breakdowns that can cause additional damage, but also eliminating unnecessary repairs that add extra costs. In both of these use cases, having the business involved in the beginning to help define the goals and also the deployment of the predictive model results is important, by creating applications that deliver the results automatically to the users.
Now, let’s think for a moment about what self-service means to the business and less-technical users, as there are two aspects to consider: predictive model creation and predictive model deployment. For predictive model creation, people often leave this to the quants of the world and, depending on the complexity of the data and use case, that is often the way to go. But today, there are low-code/no-code solutions available that can be used by less technical people. Another way to look at it is, a simple, transparent predictive model that provides a business decision superior to what’s in place today is often all that’s needed, especially for an initial model. The model can always be revised and improved in future phases with additional data, revisions to the data, and trying out various modeling techniques.
Once the models are trained, we can now consider how to deploy and operationalize the results. Business users who need to consume the predictions and take action need applications that are familiar, easy to understand, easy to interact with, and are automatically updated, whether embedded into a third-party application or standalone. When it comes to advanced analytics, true results deployment is more than data-sharing and populating a dashboard with new attributes. Business users need the model to be deployed as a scoring function that executes at run-time. This approach ensures that the predictive application and the results being delivered to the business user are up-to-date and actionable. The data changes (as it always does), the application executes and the predictive model produces timely predictions.
Please join Martha Bennett from Forrester Research and me on April 18 for a webinar to that discusses this topic and how Information Builders technology supports these different development and deployment options for systems of insight that exploit predictive analytics.